Automated table understanding using stub patterns

Abstract

Tables in documents are a rich source of information, but not yet well-utilised computationally because of the difficulty of extracting their structure and data automatically. In this paper, we progress the state-of-the-art in automatic table extraction by identifying common patterns in table headers to develop rules and heuristics for determining table structure. We describe and evaluate a table understanding system using these patterns and rules.